Four Uber rides, a $200 AI budget and a startup bought by Meta

Three stories from this week's tech news, read one at a time, look unrelated. Read together they say the same thing: companies are still spending heavily on AI, but they have started checking the receipt.

Uber: four rides in its busiest week

On its third-quarter earnings call, Uber was asked how much people actually use AI agents to book trips. The answer, as TechCrunch reported: four rides ordered through an agent during the week of Halloween, one of its busiest of October. Across Uber and Uber Eats that week there were about 320 million trips and orders. Food was a bit better, with about 500 Uber Eats orders coming through agents. DoorDash told investors it saw only a few hundred.

CEO Dara Khosrowshahi called it "very, very early and very limited traction", and then made the case that agents will still matter as one more way for customers to find Uber. He also gave the number that explains the low usage: booking a ride in the app takes about 50 seconds, and reordering a favourite meal about 90.

That's the lesson for anyone building agent products. An agent doesn't have to beat a clumsy process, it has to beat a tap that already takes under a minute. The agent ideas that work tend to be the ones attached to jobs that are slow, tedious and repeated, like reconciling invoices, triaging support tickets or migrating code. Ordering a cab is not one of them.

Apple: the daily AI allowance is shrinking

Bloomberg reported this week, per TechCrunch, that Apple cut the daily amount its engineering teams can spend on AI from $300 to $200 in October. More cuts are reportedly coming: some allowances may fall to $100 a day, and outside engineering the cap could go as low as $25. TechCrunch says it asked Apple for comment.

TechCrunch ties it to a trend earlier this year, sometimes called "tokenmaxxing", where companies pushed staff to use as much AI as possible and hoped efficiency would follow. Then the bills arrived.

AI is getting expensive at the hardware end too. Tom's Hardware has been documenting how AI demand is soaking up memory chips. In the last few weeks it has reported that the cheapest 32GB DDR5-6000 kit now costs about $450, that an RTX 5080 is selling for around $2,000 against a $999 launch price, and that Apple's manufacturing bill has climbed by $33 billion because of memory costs, with pressure expected into 2027. We've linked the articles because the details are worth reading directly.

Our read: this isn't the end of AI at work, it's the start of AI being treated like any other line on the budget. The engineers who do well will be the ones who can say what a piece of work costs in tokens and whether it was worth it.

Meta buys Conductor

Meta has acquired Conductor, TechCrunch reports, after Axios broke the news. Conductor makes a tool for running several coding agents at once, each in its own workspace, so one engineer can have different agents working on different parts of a project. The agents themselves come from labs like Anthropic and OpenAI. Conductor was founded in 2024 by Charlie Holtz, and the price wasn't disclosed.

Two practical details. Conductor's cloud service shuts down on 20 November, and the company will start deleting user cloud data after that, so if you use it, export what you need. And Holtz says to expect "something new" early next year.

Meta already has its own coding agent, Muse Code, which launched in August. Buying a company whose whole product is managing many agents at once suggests that Meta thinks the next hard problem isn't one agent writing code, it's a person supervising five of them.

Two more worth a look

Agility, the humanoid robot maker, is now public. It listed on the Nasdaq through a SPAC merger that raised $465 million. Its Digit robots mostly move heavy totes in warehouses and factories. The numbers are a good reminder of how early this is: $174,000 of revenue in the first half of 2026, a $362 million loss, and nine customers. Robotics roles are real, but this is a long bet.

Nothing says its first "AI-native product" arrives in December. CEO Carl Pei said on X that it has no new screen and "couldn't have been an app". TechCrunch notes there's no firm date and no details yet.

What this means if you're job hunting in India

• Say what the AI did, in numbers. Everyone's resume now says "built AI agents". The ones that stand out say what changed: hours saved, tickets closed, cost per task. Uber's four rides is exactly what happens when nobody checks that.
• Learn to supervise, not just prompt. Reviewing and directing several agents' work is a skill, and Meta just paid for a product built around it.
• Know what AI costs. If budgets are being capped at companies like Apple, being able to talk about cost per task is useful in interviews.
• Keep it in proportion. AI roles exist, but they're a small slice of what's posted. In the week to 4 November, 136 of the 1,614 new jobs we classified had AI or ML in the title, about 8%. We go through the rest in our weekly report.

Sources: TechCrunch (4 and 5 November 2026) and Tom's Hardware, linked above. Figures are as reported by those outlets; TechCrunch says it asked Apple for comment on the Bloomberg report.